Trajectory generation using reinforcement learning for autonomous helicopter with adaptive dynamic movement primitive

摘要

The present paper introduces a smart trajectories generation algorithm for unmanned aerial vehicles under various environments. Dynamic movement primitive is extended by adding jerk to mock the kinematics, particularly for unmanned aerial vehicles. Combining the improved dynamic movement primitive with policy learning by weighted exploration with the returns, we propose the new algorithm producing optimal trajectories under different scenarios. Furthermore, numerical simulations under several scenarios are performed, demonstrating the ability of the proposed algorithm.